DECISION FREEDOM AS A DETERMINANT OF THE ROLE OF INCENTIVE MAGNITUDE IN ATTITUDE CHANGE 1

Transcription

1 Journal of Personality and Social Psychology 1967, Vol. 6, No. 3, DECISION FREEDOM AS A DETERMINANT OF THE ROLE OF INCENTIVE MAGNITUDE IN ATTITUDE CHANGE 1 DARWYN E. LINDER, JOEL COOPER, AND EDWARD E. JONES Duke University In the forced-compliance paradigm, attitude change following a counterattitudinal performance has been shown to be both a direct (reinforcement prediction) and an inverse (dissonance prediction) function of the amount of incentive offered. An experiment successfully demonstrated that if S feels free not to comply attitude change will be inversely related to incentive magnitude, but that the positive relationship will hold if this freedom is reduced. It was hypothesized that the procedure of an earlier study by Rosenberg, whose results supported the reinforcement prediction, inadvertently reduced Ss' freedom not to comply. When this procedure was closely replicated in a 2nd experiment, the positive relationship was again found, but when the procedure was modified to make a decision not to comply a viable alternative for S, the inverse relationship resulted. The 2 experiments together show that a low incentive arouses dissonance, leading to attitude change, only when the person remains free to decide against compliance after he has been fully informed about the incentive. If the incentive is announced after the person Is committed to compliance, a reinforcement effect obtains. If a person can be induced to behave publicly in a manner that does not follow from his private attitudes, he will experience cognitive dissonance. The magnitude of dissonance will be greater when there are few reasons for complying than when there are many reasons (Festinger, 19S7). This dissonance may be reduced by an accommodating change in private attitude if other ways of reducing dissonance are not available. Thus a person who has been induced to behave in a counterattitudinal fashion will change his private attitude more the less he has been rewarded for complying. Festinger and Carlsmith (1959) found support for this proposition in a study in which subjects were persuaded (for $1 or $20) to extol the attractiveness of a dull and tedious task for the benefit of the next subject. Also, Cohen (1962) found that Yale students who were induced to write essays in favor of the New Haven police later showed more positive attitudes the smaller the incentive they had been offered to write the essay. Rosenberg (196S) has recently questioned the generality of this proposed relationship and has suggested that subjects in the experiment was facilitated by National Science Foundation Grant We are indebted to H. B. Gerard for his valuable suggestions. 245 Festinger and Carlsmith (19S9) and the Cohen (1962) experiments must have considered the incentive excessive and, because of the "evaluation apprehension" that subjects in psychology experiments commonly feel, those in the high-incentive conditions may have interpreted the experiment as one testing their honesty and autonomy, To resist influence in the face of a substantial bribe, therefore, would cause the experimenter to evaluate them favorably. Alternatively, Rosenberg suggests that the subjects may have suspected deception in the highincentive condition and angrily resisted confirming the perceived hypothesis. Either reaction might conceivably account for the obtained inverse relationship between incentive amount and degree of ultimate congruence between attitude and behavior. Rosenberg proceeded to conduct an experiment loosely replicating Cohen's (1962), the major difference being that one experimenter provided the incentive for essay writing and another measured the subsequent attitude. The two tasks were presented to the subjects as unrelated, and thus there was presumably no chance for evaluation apprehension to affect the results. Rosenberg found that attitude and behavior were most congruent when subjects were offered $5 for writing essays

2 246 D. E. LINDER, J. COOPER, AND E. E. JONES and least in line when they were offered $.50. Rosenberg's results are clearly at variance with the apparent dissonance prediction and with the findings obtained by Cohen and by Festinger and Carlsmith. Instead, the results seem to favor a reinforcement position or, as Rosenberg would prefer, a theoretical position that considers the effects of reinforcement in the context of an affectcognition consistency model. In Rosenberg's view, either the expectation or the receipt of reward strengthens and stabilizes the cognitions associated with the counterattitudinal statement the greater the reward, the greater the stabilizing effect. There is then a change in attitudinal affect in the interests of cognitive-affective consistency. Nuttin (1964) carefuly replicated Rosenberg's experiment and found that even with evaluation apprehension removed in the same manner inferred attitude change varied inversely with the amount of incentive offered. While Nuttin's results were of only borderline significance, they clearly offered more support for the dissonance proposition than for the counterproposition reflecting reinforcement theory. Aronson (1966) has criticized Rosenberg's reasoning and his conclusions on many different grounds. Perhaps his most telling criticism was that Rosenberg should have tried to reproduce the inverse incentive effects previously attributed to dissonance theory by adding conditions to his design in which the same experimenter called for the essay and measured the subsequent attitude. Aronson argued that there were many differences between the Cohen experiment and the Rosenberg replication, and to assume that his results reversed the dissonance proposition solely because evaluation apprehension was removed is unwarranted. The fact remains that Rosenberg (1965) was able to obtain a positive relationship between amount of incentive and inferred attitude change, and the intriguing empirical and theoretical problem is how to account for the fact that both dissonance and reinforcement effects have been found within the forced-compliance paradigm. Carlsmith, Collins, and Helmreich (1966) have predicted and successfully produced these opposing effects in a context approximating the original Festinger and Carlsmith (19S9) study. When the subject was induced to describe the task as attractive to the next "unsuspecting subject," the former's subsequent taskattractiveness ratings were more positive in low- than in high-incentive conditions. When the subject was instead asked to write an essay praising the task, portions of which might later be used by the experimenter, rated attractiveness varied directly with the amount of incentive offered. Carlsmith et al. argued that amount of incentive will relate directly to attitude change (a reinforcement effect) whenever the dissonance involved in a counterattitudinal act is minimal. The subject who complied in the essay-writing conditions of their experiment had a number of legitimate reasons for doing so, and the experimenter, the only person to read the essay, knew full well that the essay did not express the subject's private opinion. Dissonance should be much greater in the conditions requiring the subject to dupe another person like himself. The Carlsmith et al. (1966) experiment is especially important because of the care with which it was conducted, the clear replication of the Festinger and Carlsmith (19S9) results it provides, and the separate elicitation of both dissonance and reinforcement effects within the same general design. But while they may account for the Rosenberg reinforcement effect, Carlsmith et al. are left without any clear explanation for Cohen's (1962) results. After all, he required an essay rather than a deceitful confrontation with another subject and obtained dissonance rather than reinforcement effects. One could argue that attitudes toward the New Haven police are likely to be more central and important than attitudes toward a boring task, and thus a counterattitudinal essay is more inherently dissonant in the former case. Or, one could argue that the subjects in Cohen's experiment were not really assured anonymity (as in Carlsmith et al.'s). Nevertheless, the empirical discrepancies existing in the forced-compliance literature are not entirely reconciled by the Carlsmith et al. stud}'. The major focus of these studies has been the relationship between the amount of incentive offered and subsequent attitude

3 change, but a clear prediction from dissonance theory cannot be made unless the subject makes his decision to comply ajter considering the incentive magnitude. The incentive must be one of the conditions potentially affecting the decision to comply rather than a reward for having already so decided. Both Cohen (1962) and Rosenberg (1965) reported that they took care to assure subjects that the decision to write the essay was entirely their own. It may be argued, however, that Rosenberg's major alteration of Cohen's procedure, the separation of the compliant-behavior setting from the attitudemeasurement setting to eliminate evaluation apprehension, reduced his subjects' freedom not to comply. When Rosenberg's subjects arrived for the experiment, they found him busily engaged and were given the option of waiting for "IS or 20 minutes" or, as an afterthought, participating in "another little experiment some graduate student in education is doing." Professing to know little about this other experiment except that it "has to do with attitudes" and "I gather they have some research funds," Rosenberg did not pressure the subject into a decision, but let him decide for himself whether he wanted to participate or wait. Having made the decision to participate, each subject further strengthened his commitment by walking to the location of the second experiment. The choice then offered by the second experimenter was considerably less than a free one. Being already effectively committed, the subject would be more likely to treat the subsequent monetary offering as a bonus for prior compliance than as one of the conditions to be considered in making a free choice. If the preceding argument is correct, Rosenberg's findings cannot be compared with Cohen's because different conditions prevailed in the two experiments when the counterattitudinal essays were written. Rosenberg inadvertently made it difficult for subjects not to comply and found that degree of attitude change was positively related to incentive magnitude, in support of a reinforcement position or an affective-cognitive consistency model (Rosenberg, 1960). In contrast to this, Cohen's procedure presented DECISION FREEDOM AND ATTITUDE CHANGE 247 the choice not to comply as a more viable alternative and found that attitude change was inversely related to incentive magnitude, in support of a derivation from dissonance theory, A meaningful resolution of these discrepant findings would be to show that the effects of incentive magnitude on attitude change are either direct or inverse, depending on the presence or absence of freedom not to comply. The first experiment to follow was conducted as a direct test of the role of such freedom to choose not to engage in counterattitudinal behavior. Method EXPERIMENT I Attitude issue and subjects. At the time of the first experiment a rather heated controversy was raging in the state of North Carolina concerning the wisdom of a law that forbade Communists and Fifth Amendment pleaders from speaking at statesupported institutions. On the basis of informal opinion sampling, fortified by the plausible expectation that students deplore implied restrictions on their own freedom to listen, we assumed that college-student subjects would be strongly opposed to speaker-ban legislation. The issue thus seemed comparable to "the actions of the New Haven police" (Cohen, 1962) and to a ban on Ohio State's participation in the Rose Bowl (Rosenberg, 1965). Fifty-five introductory psychology students at Duke University served as subjects in the experiment. Forty subjects (IS males and 25 females) were randomly assigned to four experimental conditions 2 ; 13 were subsequently assigned to a control condition. All experimental subjects were asked to write a "forceful and convincing essay" in favor of the speaker-ban law. After writing the essay, each subject was asked to indicate his opinion about the speaker-ban law by checking a point on a 31-point scale comparable to Cohen's (1962) and Rosenberg's (1965) measure. The scale read, "In my opinion the Speaker Ban Law of North Carolina is...," followed by 31 horizontal dots with seven labels ranging from "not justified at all" to "completely justified." Subjects in the control condition merely filled out the scale without having previously written a pro-speaker-ban essay. Procedure and Design The basic procedure was closely modeled after that of Cohen (1962) except that the subjects were recruited from the introductory psychology course 2 Two more experimental subjects were actually run whose data were not analyzed. One of these was obviously in favor of the speaker-ban law at the outset, and the other was the victim of experimenter error in presenting instructions.

4 248 D. E. LINDER, J. COOPER, AND E. E. JONES and came individually to the laboratory, rather than being approached in their dormitory rooms. The experimenter introduced himself as a graduate student in psychology. In the free-decision condition he immediately said, "I want to explain to you what this task is all about. I want to make it clear, though, that the decision to perform the task will be entirely your own." In the no-choice condition he merely said, "I want to explain to you what this task that you have volunteered for is all about." He then proceeded in both conditions to provide the following rationale for the essay-writing task: The Association of Private Colleges of the Southeast, of which Duke is a member, is considering the adoption of a uniform speaker policy that would be binding on its member schools. Before they can decide what kind of policy to adopt, if indeed they decide to adopt one, they have undertaken a large scale research program in order to help them understand what the issues really are. This study is part of that program. The APCSE is working through the Department of Psychology here at Duke and through the departments of psychology at other private schools in the area because of the access which the department has to a wide cross-section of students such as yourself who must participate in psychological experiments and because of the number of graduate students that are available to conduct research. We have found, from past experience, that one of the best ways to get relevant arguments on both sides of the issue is to ask people to write essays favoring only one side. We think we know pretty much how you feel about the student's rights in this matter. [Here the experimenter paused and waited for a comment that would confirm the subject's initial opinion opposing the speaker-ban law. Only one subject expressed a favoring opinion at this point; see Footnote 2.] Nonetheless, what we need now are essays favoring the speaker ban. At this point, the free-decision and no-choice conditions again diverged. In the free-decision condition the subjects were told that the APCSE was paying $.50 (low incentive) or $2.50 (high incentive) in addition to the standard experimental credit given to all subjects. The experimenter again stressed that the decision to write the essay was entirely up to the subject and that he would receive experimental credit in any case. In the no-choice condition the experimenter acted as if, naturally, the subject in volunteering for the experiment had committed himself to its requirements. He simply pointed out that the experiment involved writing a strong and forceful essay favoring the speakerban law. After the subject was handed a pencil and some paper, but before he began to write, the experimenter broke in: "Oh yes, I almost forgot to tell you,... The APCSE is paying all participants $.50 [or $2.50] for their time." In all conditions subjects were paid, before they wrote the essay, the amount of money promised them. The experimenter then left the room and allowed the subject 20 minutes to complete his essay. When he returned, the experimenter collected the essay, administered the brief attitude scale, and interviewed the subject concerning his perceptions of the experiment. No subject indicated any suspicion regarding the true purpose of the experiment. The purpose was then explained to each in detail, and all deceptions were revealed. None of the subjects recalled having any doubts about the existence of the fictitious APCSE. Each subject was ultimately allowed to keep $1.50. Because they were made to realize that they were assigned by chance to the high-inducement condition, those who had initially received $2.50 were quite agreeable when asked to return $1 of their monej'. Subjects in the lowinducement condition were delighted to learn of their good fortune that they would receive $1 more than they had bargained for. Results Before the results bearing on the central hypothesis are presented, it is of interest to note the difference in decision time in the two free-decision conditions. After the experimenter began to notice that free-decision subjects in the low-incentive condition took much longer to make up their minds about writing the essay than free-decision subjects in the high-incentive condition, he started to record decision times with a hidden stopwatch. The last seven subjects in the lowincentive condition took an average of seconds to reach a decision; the comparable mean for the last seven subjects in the high-incentive condition was seconds. In spite of the reduced n, this difference is significant (p <.025, U test). This evidence strongly suggests that there was greater predecisional conflict in the low-incentive condition, and thus the conditions are appropriate for testing the dissonance hypothesis: since predecisional conflict leads to postdecisional dissonance (Festinger, 1964), more dissonance and hence more attitude change should occur in the free-decision-low-incentive condition. After establishing that the means for female and male subjects were nearly identical ( =.18), the posttreatment attitude scores were placed in a simple 2 (for Degree of Decision Freedom) X 2 (for Level of Incentive) factorial design. The means for each condition are presented in Table 1. Scale values could range from 1.0 (antispeaker ban) to 7.0 (prospeaker ban). Table 2 summarizes the analysis of variance and appro-

5 DECISION FREEDOM AND ATTITUDE CHANGE 249 TABLE 1 ATTITUDE-SCALE MEANS OBTAINED IN THE FIVE CONDITIONS : EXPERIMENT I No choice Free decision Control 15 $ » 2.96 Incentive $ Note. n = 10 under both incentives for free-choice and free-decision conditions. For the control condition, n 13. a The higher the number, the more the speaker-ban law was considered to be justified. b Since subjects in the control condition were all run after the main experiment was completed, the mean for this condition is presented only as an estimate of student opinion toward the issue in the absence of dissonance or incentive effects. The data from the control condition were not included in the statistical analysis. priate orthogonal comparisons. The prediction that the amount of inferred attitude change would relate positively to inducement level in the no-choice conditions and negatively in the free-decision conditions is clearly confirmed (^"i, so = 8.70; p<.ql). The dissonance effect in the free-decision condition was itself significant; the reinforcement effect in the prior-commitment condition was not. The control subjects, who checked the scale without writing an essay, were about as much against the speaker ban as subjects in the conditions where little or no change was predicted. In an effort to shed light on possible mechanisms underlying these findings, the essays themselves were examined. The average number of words per essay was 192,3, and there were no significant differences among the four conditions in essay length. The essays were evaluated in a manner sirni- TA13LE 2 SUMMARY OF ANALYSIS OF VAETANCE : EXPERIMENT I Source of variation Choice (A) Incentive (B) AXB Error Low incentive vs. high incentive within free-decision conditions Low incentive vs. high incentive within no-choice conditions MS F <1 <1 8.70**** 7. 57**** 2.01* Note. Two-way analysis of experimental condition:;. * f <.20, df = 1/36. ****/> <.01, rf/ = 1/36. lar to that described by Rosenberg (196S), Two independent raters, blind as to the subject's condition, rated the essays in terms of the degree of organization manifested and the degree of "intent to persuade." Each of these ratings was made in terms of a 5-point scale. The judges agreed or were 1 point discrepant in 72% of the organization ratings and 85% of the persuasiveness ratings. These percentages of agreement were comparable to those obtained by Rosenberg (1965), but two independent judges using S-point rating scales should, by chance, be no more than 1 point discrepant on more than 50% of their ratings. When a more traditional estimate of the reliability of the ratings was calculated (Winer, 1962, pp. 124 ff.), it was found that the reliability coefficient for the ratings of degree of organization was.54, and the coefficient for the ratings of persuasiveness was.55. These coefficients estimate the reliability of the ratings that result from averaging over the two judges. When these ratings were submitted to an analysis of variance, there were no differences among conditions in either organization or persuasiveness. Since the reliability of the ratings discussed above was quite low, an attempt was made to obtain ratings of acceptably high reliability. Two varsity debate partners agreed to rate the essays. General criteria to be used in determining the ratings were discussed, but the ratings were made independently. Each essay was rated for the persuasiveness of the presentation on a 7- point scale, Sixty percent of these ratings were no more than 1 point discrepant; the reliability coefficient was,48. (The chance percentage for agreements or 1-point discrepancies is 39% when a 7-point scale is used by two independent judges.) There were again no differences among the conditions in the rating received. Also, no between-condition differences appeared on the ratings made by any individual judge. Discussion The major purpose of the present experiment was achieved: to show that dissonance and reinforcement effects can be obtained within the same forced-compliance paradigm

6 250 D. E. LINDEE, J. COOPER, AND E. E. JONES by varying the degree to which the subject is committed to comply before learning about the monetary incentive. Subjects who commit themselves after weighing the unpleasantness of the essay-writing task against the amount of incentive offered show the effects predicted by dissonance theory. The decision-time data strongly suggest that the subjects do in fact consider the essaywriting task unpleasant. Subjects who are not free to decide against compliance and then learn about a financial "bonus" produce results in line with reinforcement theory (that which is associated with something of value itself takes on value) or in line with the more complex affective-cognitive consistency model espoused by Rosenberg. The present study was stimulated by Rosenberg's (1965) experiment, but the relevance of the results to a critique of Rosenberg's conclusions rests on the claim that his way of removing evaluation apprehension precommitted the subject to an unpleasant task before he had a chance to weigh the incentive for compliance. If this criticism is valid, then it should be possible to reproduce Rosenberg's results by closely replicating his procedures, and to obtain the converse of these results (confirming the dissonance prediction) by insuring that the subject does not commit himself before being confronted with the incentive for compliance. A second experiment was planned in an attempt to do precisely this. Method EXPERIMENT II Attitude issue. As we prepared to run the second experiment, certain paternalistic policies of the Duke University administration were being challenged by the undergraduates, and there was a movement toward liberalization of in loco parenlis social regulations. It was assumed, therefore, that undergraduates who were induced to write forceful and convincing essays in support of strict enforcement of in loco parenlis policies would be performing a counteratlitudinal task. Subjects. Fifty-nine male introductory psychology students volunteered to participate for experimental credit in a study described as an "Altitude Survey." The data of SO of these students, who were randomly assigned to the four experimental conditions and the control condition, were used in the reported analysis. Six subjects were eliminated because they did not complete the experimental procedure. Usually, they chose to read or study while waiting for the first experimenter rather than go to the second experimenter. Another subject was eliminated because he was initially in favor of strict in loco parenlis policies, and writing the essay would not have been counterattitudinal for him. Only two subjects who had completed the procedure were eliminated from the analysis. The first of these was excluded when it was discovered during the final interview that he had misinterpreted the attitude questionnaire. The second was eliminated because he accurately perceived the true purpose of the experiment. Both subjects had been assigned to the jree-decision-high-incentive condition. The results of the study are not changed if these two subjects are included in the analysis. Procedure and design. The procedure was a close approximation to that used by Rosenberg (196S). All subjects reported to the office of the first experimenter (Ei) where they found Ei engaged in conversation with another student and were told, "I'm sorry, but I'm running late on my schedule today, and I'll have to keep you waiting for about IS or 20 minutes. Is that all right?" All subjects agreed to wait. Each experimental subject was then told: Oh, I've just thought of something; while you're waiting you could participate in another little experiment that some graduate student in education is doing. This fellow called me the other day and said he needed volunteers in a hurry for some sort of study he's doing I don't know what it's about exactly except that it has to do with attitudes and that's why he called me, because my research is in a similar area as you'll see later. Of course, he can't give you any credit but I gather thay have some research funds and they are paying people instead. So, if you care to do that, you can. At this point, one-half of the experimental subjects (prior-commitment condition) were allowed to leave for the second experiment without further comment by Ei. Since it was believed that Rosenberg's procedure restricted subjects' freedom not to comply with the task of the "little experiment," it was decided to manipulate degree of choice by removing this restriction. Thus, for subjects in the freedecision condition, after the subject had agreed to participate in the second experiment, Ei added: All I told this fellow was that I would send him some subjects if it was convenient but that I couldn't obligate my subjects in any way. So, when you get up there, listen to what he has to say and feel free to decide from there. All experimental subjects then reported to the second experimenter (»). To control for the effects of experimenter bias, En was not informed whether the subject was in the prior-commitment condition or the free-decision condition. 2 presented himself as a graduate student in the Department of Education and introduced the essay-writing task using a pro-

7 DECISION FREEDOM AND ATTITUDE CHANGE 251 cedure that, as in Experiment I, very closely approximated Cohen's (1962). Rather than the free-decision versus no-choice manipulation of Experiment I, & began by saying to all subjects, "At the present time, Duke University is beginning to question the wisdom of assuming the role of 'substitute parent' to its students." From that point, the instructions paralleled those of Experiment I with the substitution of in loco parentis regulations for the speaker-ban law. After confirming that the subject held an opinion opposed to rigid in loco parentis regulations,» concluded: What we need now are essays favoring a strict enforcement of in loco parentis. So, what we would like you to do if you are willing 3 is to write the strongest, most forceful and most convincing essay that you can in favor of a strict enforcement of the substitute parent concept here at Duke. It was then explained that the sponsoring agency was offering either $.50 (low-incentive conditions) or $2.50 (high-incentive conditions) for participation in the study. When the subject agreed to write the essay, he was paid the money promised to him and then began the task. After completing the essay and being thanked and dismissed by E?, all experimental subjects returned to Ei's office. To introduce the dependent measure, Ei explained: What I had wanted you to do was participate in a continuing study I carry on every semester as a sort of Gallup poll to keep a check on opinion patterns on different University issues. I'd like you to fill out this questionnaire as an objective indication of your opinions and when you've finished I'd like to chat for a while about various issues on campus. OK? ET. was not informed of the amount of money the subject had received, and in no case did he find out until after the subject had completed the dependent measure. The dependent measure consisted of an eightitem questionnaire dealing with various university issues. The critical item read, "How justified is the Unversity's policy of assuming parental responsibilities for its students?" and was accompanied by the familiar 31-point scale. When the subject had completed the questionnaire, Ei put it aside (without looking at the responses) and began a structured interview that included probes for suspicion and checks on perceptions of the manipulations. When Ei was satisfied that the subject had not perceived the true purpose of the experiment, 3 This vague statement of choice was given to all subjects in order to keep the instructions constant across experimental groups and to enable & to remain "blind" as to the condition of each subject. It was assumed that the crucial manipulation of free decision versus prior commitment had already been accomplished by Ei. he revealed the deceptions and explained the necessity for them. As in Experiment I, all experimental subjects agreed to accept $1.50 for their time. Subjects assigned to the control condition also found Ei engaged in conversation and were asked if they could return in 15 or 20 minutes. Upon their return they were treated exactly as experimental subjects. These procedures resulted in five conditions: two levels of incentive magnitude under a condition of free decision, the same two levels under a condition of prior commitment, and the control condition. Results The mean attitude-scale scores on the critical item for each of the five conditions are presented in Table 3. It can be seen that the results are very similar to those of Experiment I. The data were submitted to a one-way analysis of variance, summarized in Table 4. The overall treatment effect was significant (F 4i «= 4.02; p <.01). The two comparisons reflecting the hypotheses of this study were also significant: Within the freedecision conditions a low incentive produced more inferred attitude change than a high incentive (F 1>45 =6.82; p <.025). Within the prior-commitment conditions this effect was reversed, and a high incentive produced more inferred attitude change than a low incentive (F li43 = 4.90; p <.OS). The position of the control group indicates that differences between the experimental conditions resulted from positive attitude change rather than a combination of positive and negative changes. Once again we attempted to investigate the possibility that these effects were mediated by some aspect of the counterattitudinal performance. Two raters, working independently and without knowledge of the experi- TABLE 3 ATTITUDE-SCALE MEANS OBTAINED IN THE FIVE CONDITIONS : EXPERIMENT II Prior commitment Free decision Control $ " 3.64 Incentive 2, Note. n = 10 in all conditions. a The higher the number, the more strict application of in loco parentis regulations was considered justified.

8 252 D. E. LINDER, J. COOPER, AND E. E. JONES Treatment Error TABLE 4 OP ANALYSIS OP VARIANCE: EXPERIMENT II Source of variation* 1 Lo\v incentive vs. high incentive free-decision conditions Low incentive vs. high incentive prior-commitment conditions.1/ F 4.02**** 6.82*** 4.90** Note. One-way analysis of five conditions. :1 The control condition differs significantly from both the prior-commit men t-high-inceiitive and thefree-dedsion-lowincentive conditions. **/> <.OS, ilf = 1/45.. ***/> <.025, rf/ 1/45. ****/> <.01, <// = 4/45. mental conditions, rated each essay on 7- point scales for the extremity of attitudinal position advocated, the persuasiveness of the essay, and its degree of organization. The two raters agreed or were within 1 point of each other for 65% of the essays when estimating the attitudinal position, 60% when rating them for persuasiveness, and 52.5% when rating them for organization. The reliability coefficient for the estimated attitudinal position (Winer, 1962, pp. 124 ff.) was.67, the coefficient for the persuasiveness ratings was.51, and the coefficient for the organization ratings was.38. There were no differences among conditions on any of these ratings. The essays were then rated for the persuasiveness of the presentation on a 7- point scale by the same varsity debaters as had rated the essays from Experiment I. The debaters agreed or were within 1 point of each other for 65% of the essays, and the reliability coefficient was a somewhat more acceptable.71. Again, however, there were no differences among conditions on these ratings, whether the judges' ratings were averaged or each judge's ratings were examined separately. In a final attempt to find a performance difference among the conditions the number of words in each essay was counted; the conditions were compared on this measure of performance and were found not to differ from one another. Discussion The results of Experiment II support the argument that Rosenberg's (1965) procedure for the elimination of evaluation apprehension committed his subjects to perform the essay-writing task before they learned of the nature of the task and the amount of reward offered. The positive relationship between incentive magnitude and attitude change in the prior-commitment conditions of the present experiment replicates the no-choice results of Experiment I and the relationship found by Rosenberg (1965). It could be argued on this basis alone that such procedures as Rosenberg's have the same effect as allowing the subject no choice concerning performance of the counterattitudinal act. The argument becomes much more convincing, however, if it can be shown that appropriate alteration of Rosenberg's procedure, reducing the prior commitment of the subject, leads to an inverse relationship between incentive magnitude and attitude change. The free-decision conditions of Experiment II demonstrate precisely this point: when the subject does not feel that he has previously committed himself to performance of the counterattitudinal action requested by E- 2, attitude change is an inverse function of incentive magnitude. It should be noted here that a "balanced replication" (Aronson, 1966) of Rosenberg's (1965) study was required. Had Experiment II included only the free-decision conditions it would be possible to argue that our procedure was not successful in eliminating evaluation apprehension and that the results reflected once again the effect of this contaminant in research on forced compliance. The results of the prior-commitment conditions of Experiment II, however, counter this criticism. A persistent critic might still argue that the free-decision manipulation reintroduced evaluation apprehension. Perhaps the comment added to create the free-decision condition in some way increased the chances that subjects would see the experiments as related, However, the structured interview conducted by 1 revealed no differential level of suspicion between the priorcommitment and free-decision conditions. In the absence of a reliable and valid measure of evaluation apprehension, we can do no more than contend that our interview was sensitive enough to detect suspicion and that

9 DECISION FREEDOM AND ATTITUDE CHANGE 253 we found no indication of differential suspicion among the conditions. The results of the two studies reported above imply that the discrepancy between Cohen's (1962) findings and the results of Rosenberg's (196S) experiment may indeed be resolved in the manner indicated earlier in this paper. For Cohen's subjects the decision not to comply was a viable alternative at the time they were confronted with the essay-writing task and offered an incentive of certain value. Under such conditions dissonance will be induced whenever the incentive is not large enough to justify performance of the task, and incentive magnitude will be inversely related to subsequent attitude change. However, if a subject's freedom not to comply has been restricted before he is confronted with the task and with a clear description of the incentive, dissonance cannot be induced by an incentive of insufficient magnitude. Under these conditions, the reinforcing properties of the incentives will lead to a positive relationship between incentive magnitude and attitude change. Although Rosenberg (1965) demonstrated such a relationship, his assertion that it may be obscured by failure to remove evaluation apprenhension seems no longer tenable. No attempt was made in the procedure of Experiment I to remove evaluation apprehension, and yet the results are very similar to the results of Experiment II. Rosenberg (1966) has more recently advanced two additional hypotheses intended to resolve discrepancies in the forced-compliance literature. The first of these is that we must distinguish counterattitudinal actions that are simple and overt from those featuring the elaboration of a set of arguments. Supposedly a performance of the former kind (e.g., eating a disliked food) will lead to the inverse relationship between attitude change and incentive magnitude, while an act of the latter kind (e.g., writing a counterattitudinal essay) will result in a positive relationship. The second hypothesis is that we must distinguish between two kinds of counterattitudinal performances: (a) those carried out under instructions that lead the subject to believe his performance will be used to deceive others, and (b) those following from instructions to elaborate, for some reasonable and legitimate purpose, a set of arguments opposite to his private opinion. It is hypothesized that even if the actual task is the same, say essay writing, the first type of instruction will lead to an inverse relationship between incentive magnitude and attitude change, and the second type of instruction will lead to a positive relationship. In the studies reported above the subject's task was presented with no hint that his performance would be used to deceive anyone, and the task in all cases was to elaborate a set of arguments opposite to his own opinion. It follows from the two hypotheses suggested by Rosenberg (1966) and presented above that we should not have been able to obtain the inverse relationship between incentive magnitude and attitude change using our procedures. However, in both experiments, we obtained the positive and the inverse relationship. We are forced to conclude that neither the "simple versus complex" hypothesis nor the "duplicity versus legitimate" hypothesis can account for the present results. In place of these hypotheses we conclude that at least some of the discrepancies in the forced-compliance literature may be resolved by closer attention to the role of decision freedom at the time the incentive is offered. A barely sufficient incentive for making counterattitudinal statements does result in dissonance and subsequent attitude change if the subject feels he is quite free not to comply. When the freedom not to comply is removed or markedly decreased, on the other hand, attitude change is greater the greater the incentive for compliance. REFERENCES ARONSON, E. The psychology of insufficient justification. In S. Feldman (Ed.), Cognitive consistency: Motivational antecedents and behavioral consequents. New York: Academic Press, Pp CARLSMITH, J. M., COLLINS, B. E., & HELMREICH, R. L. Studies in forced compliance: I. The effect of pressure for compliance on attitude change produced by face-to-face role playing and anonymous essay writing. Journal of Personality and Social Psychology, 1966, 4, COHEN, A. R. An experiment on small rewards for discrepant compliance and attitude change. In

10 254 D. E. LINDER, J. COOPER, AND E. E. JONES J. W. Brehm & A. R. Cohen, Explorations in cognitive dissonance. New York: Wiley, Pp FESTINGER, L. A theory of cognitive dissonance, Evanston, 111.: Row, Peterson, FESTINGER, L. Conflict, decision, and dissonance. Stanford: Stanford University Press, FESTINGER, L., & CARLSMITH, J. M. Cognitive consequences of forced compliance. Journal of Abnormal and Social Psychology, 1959, 58, NUTTIN, J. M., JR. Dissonant evidence about dissonance theory. Paper read at 2nd conference of experimental social psychologists in Europe, Frascati, Italy, (Miraeo) ROSENBERG, M. J. An analysis of affective-cognitive consistency. In C. I. Hovland & M. J. Rosenberg (Eds.), Attitude organization and change,. New Haven: Yale University Press, Pp ROSENBERG, M. J. When dissonance fails: On eliminating evaluation apprehension from attitude measurement. Journal oj Personality and Social Psychology, 1965, 1, ROSENBERG, M. J. Some limits of dissonance: Toward a differentiated view of counter-attitudinal performance. In S. Feldman (Ed.), Cognitive consistency: Motivational antecedents and behavioral consequents. New York: Academic Press, Pp WINER, B. J. Statistical principles in experimental design. New York: McGraw-Hill, (Received August 1, 1966)